Survey weights in the ESS
For data collection the ESS uses strictly probability-based samples. Every element in the ESS target population should therefore have a greater than zero probability of being included into the sample. When analysing ESS data estimates, the likelihood of each respondent to be part of the sample should also be taken into account - which means that the most accurate estimates will be obtained only after weighting the data.
Which weighting variables are there?
Two weighting variables are available: Design weights (dweight) and Post-stratification weights (pspwght).
Several countries use complex sampling designs where some groups or regions of the population have higher probabilities of selection. The main purpose of the design weights is to correct for the fact that in some countries respondents have different probabilities to be part of the sample due to the sampling design used. Applying the weights allows for the construction of design unbiased estimators. The design weights are computed as the inverse of the inclusion probabilities, i.e. the probability of each person to be included into the sample. The inverse inclusion probabilities are then scaled such that their sum equals the net sample size and the mean equals one.
For further information on design weights, see also the ESS Data Documentation Reports.
While the design weights account for differences in inclusion probabilities, sampling errors (related to attempting to measure only a fraction of the population) and possible non-response errors (which may lead to a systematic over- or under-representation of people with certain characteristics) are still present. Post-stratification weights are a more sophisticated weighting strategy that uses auxiliary information to reduce the sampling error and potential non-response bias. They have been constructed using information on age group, gender, education, and region. The post-stratification weights are obtained by adjusting the design weights in such a way that they will replicate the distribution of the cross-classification of age group, gender, and education in the population and the marginal distribution for region in the population. The population distributions for the adjusting variables were obtained from the European Union Labour Force Survey.
For a short description of how the post-stratification weights were computed, see the documentation. For additional details, see the reports listed in the sidebar.
Using the ESS survey weights
It is highly recommended to consult the ESS weighting guide before conducting any analysis of ESS data, especially if:
- data from more than one country and round is pooled for the analysis;
- model-based inference is used (e.g. linear regression models, generalised regression model, Bayesian methods, etc.).